Hi,
On 14 Oct 2008, at 14:35, Michael W. Cole wrote:
> Hi Steve,
>
> Thank you for the clarification. My purpose for modeling each run
> separately was (perhaps naively) to remove the inter-run mean signal
> variance from the data. Is there any way to de-mean the runs before
> (or during) combining them with FE modeling?
I'm not sure whether you're asking about the cross-run variance or
signal means. All timeseries means are removed at first level anyway.
Then, at first level, each run will generate a contrast strength
(COPE) for each first-level contrast, which tells you (for example)
how strong activation is above baseline level, or compared with a
different activation condition. Then when you do a second-level
average across runs for a given subject, when you select the FE option
the cross-run variability in the COPE value is ignored (and thrown
away). Effectively it's just an averaging, using the first-level
variance to weight across runs. So I'm not exactly sure what you're
thinking, but I'm fairly sure that this is doing what you want.
Cheers.
>
>
> Thanks,
> Michael
>
> On Tue, Oct 14, 2008 at 3:25 AM, Steve Smith <[log in to unmask]>
> wrote:
> Hi - at second level, with the FE option, you do not need to include
> one EV per run, as the FE modelling is not trying to model the cross-
> run variance anyway, it is just pooling across the first-level
> variances (that's the definition of a FE model). So if you do this
> (i.e., following the example from the FEAT manual) all should work ok.
>
> Cheers.
>
>
>
>
> On 13 Oct 2008, at 21:43, Michael W. Cole wrote:
>
> Hello,
>
> I'm using FEAT to run a 3-level analysis. My study has 10 runs per
> subject with 15 subjects. The second level combines runs (using
> fixed-effects) and the third level combines subjects (using mixed-
> effects). I included a separate EV for each run at the second level
> to remove inter-run mean effects (making it repeated-measures). I
> tried doing this (one EV per subject) at the 3rd level, using FLAME,
> and I get this error: "Singular design. Number of EVs > number of
> time points." and the script crashes. It works, however, if I
> perform a fixed-effects analysis instead of using FLAME.
>
> Is there a better way to account for run and subject differences in
> these models? Why would this work with fixed-effects but not FLAME?
>
> Thank you,
> Michael
>
>
> --
> Michael W. Cole
> Ph.D. candidate, Center for Neuroscience
> University of Pittsburgh
>
>
> ---------------------------------------------------------------------------
> Stephen M. Smith, Professor of Biomedical Engineering
> Associate Director, Oxford University FMRIB Centre
>
> FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
> +44 (0) 1865 222726 (fax 222717)
> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
> ---------------------------------------------------------------------------
>
>
>
> --
> Michael W. Cole
> Ph.D. candidate, Center for Neuroscience
> University of Pittsburgh
---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre
FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------
|